Use case
An audit trail for autonomous agents
When an agent acts on its own, someone eventually asks why. MADB makes that answer a query, not a forensic project.
An agent that acts and leaves no trail is an unauditable actor. MADB records each decision with its causes, so an agent audit trail exists by default — not reconstructed after an incident.
Replay, don’t reconstruct
When finance questions an auto-issued refund, the agent replays the exact chain that led to it with trace_cause — back to the customer’s original dispute. Nothing is inferred after the fact.
Built for review
- Every decision links to the evidence that drove it
- Lineage is captured as the agent works, not bolted on later
- Memory stays local and yours
Enforced governance and tamper-evident audit are the paid layer on the roadmap; the causal record they build on is live today.